Browser Use MCP server

The Browser Use MCP server gives AI agents the Browser Use browsing agent, either as a local open-source server with direct browser tools or as a cloud server that runs whole web tasks.

Checked by CookiesForAI on : repository facts (stars, last push, archived) re-read from GitHub; listing in the MCP Registry checked.

Written by

Connect Browser Use

Claude Code, cloud server

Terminal
claude mcp add -t http -H "x-browser-use-api-key: YOUR_API_KEY" browser-use https://api.browser-use.com/v3/mcp

Cursor or Claude Desktop, cloud server

Config
{"mcpServers": {"browser-use": {"url": "https://api.browser-use.com/v3/mcp", "headers": {"x-browser-use-api-key": "YOUR_API_KEY"}}}}

Claude Code, local open-source server

Terminal
claude mcp add browser-use -- uvx --from 'browser-use[cli]' browser-use --mcp

Get a cloud API key at cloud.browser-use.com. For the local server, set OPENAI_API_KEY or ANTHROPIC_API_KEY in its environment, and in Claude Desktop use the full path to uvx if the app cannot find it.

Or paste this into your coding agent: Connect the Browser Use MCP server to my coding agent using its documented setup: https://github.com/browser-use/browser-use. Checked by CookiesForAI on : repository facts (stars, last push, archived) re-read from GitHub; listing in the MCP Registry checked.

Your first question with this server installed

1,162tokens in the tool list, sent to the model with each question

Browser UseTool list 1,1627 tools listed without a key

Typical serverTool list 1,162 (median of 13 servers)

Browser Use's tool list takes 1,162 tokens. That is close to the median tool list of the 13 servers we measured on (1,162 tokens).

Measured without a key or sign-in. Assistants that load every tool send this list to the model with each question; some load tools only when a task needs them. Tokens are counted with OpenAI's o200k tokenizer; other models count a little differently.

Measured . How we measure.

What the Browser Use MCP server does

The Browser Use MCP server connects an AI agent to Browser Use, an open-source library in which a language model operates a real Chrome browser to finish a task. It comes in two forms. The cloud server takes a whole task in plain words, such as finding a flight or filling in an application, runs it in Browser Use's cloud browsers, and reports status, cost, a live view URL and the result, optionally shaped by a JSON schema.

The local server runs on your machine through uvx and gives the agent lower-level tools: navigate, click and type by element index, read the page state, scroll, manage tabs, extract content, get HTML and take screenshots. When direct control fails, a fallback tool hands the whole task to the Browser Use agent. Cloud sessions can keep the browser alive for follow-up tasks and use saved browser profiles for sites that need a login.

The browser-use repo on GitHub, which holds the local MCP server, had 116,895 stars on . Those stars are for the whole library, not for the MCP server alone.

Tools it gives your agent

  • run_session
  • get_session
  • send_task
  • stop_session
  • list_sessions
  • list_browser_profiles
  • browser_navigate
  • browser_click
  • browser_type
  • browser_get_state
  • browser_extract_content
  • retry_with_browser_use_agent

When to use Browser Use

  • You want to hand off a whole web chore, like comparing prices on several shops, and get back the result.
  • You need a browser task to run in the cloud while your own machine stays free.
  • You already use the Browser Use library and want the same agent inside your editor.
  • A task needs a logged-in session that you saved as a browser profile.

When to pick something else

  • Cost control without watching: each run spends model tokens plus browser time, and the local server needs your own LLM API key.
  • Step-by-step deterministic testing; Playwright MCP gives exact, repeatable actions.

What Browser Use needs

  • Cloud: a Browser Use Cloud API key
  • Local: Python with uv, Chrome, and an OpenAI or Anthropic API key

What the Browser Use MCP server costs

The local server is free and open source under MIT and uses your own LLM API key; the cloud server is pay as you go, with free signup credit for eligible accounts.

Current prices: browser-use.com . We do not list prices because vendors change them often. Check the publisher's own page.